scaleutils: add nodeselector package to select nodes suitable for termination

[jrasell]

closes #417

This work allows operators to specify the strategy which is used to select nodes for termination during a scale in action. Up until now, the Nomad Autoscaler would simply pick nodes from the top of the list as returned from the Nomad node list API endpoint. This method is still included and is now known as newest_create_index.

In order to explicitly configure the selector behaviour, a new policy target configuration option node_selector_strategy has been made available. This config option is parsed by the scaleutils helper and therefore is available to all cluster target plugins which utilise it.

In the absence of an operator configured option, the value defaults to least_busy. This strategy calculates the percentage allocation of each node in the pool, selecting those which have the lowest value and therefore are allocated the least amount of resources. The resources are CPU and memory.

The second selector strategy available is empty. This strategy will only select nodes for termination that do not have non-terminal allocations. This is ideal for worker pools running batch workloads which should not be interrupted or migrated.

Admin notes:

• Documentation on the Nomad site will be required as a follow up step.
• To those internally testing this, I have some new TF I have been working on that will allow easier testing of different architectures such as this example. I will share this privately for now as it is very alpha.
• This PR only implements a single strategy whereas the original issue states several. I will raise follow up issues to individually track additional selector strategies. This is ideal candidates for first-issues and will be marked as such.

[cgbaker]

one array bounds question and a few comments 👍

[cgbaker]

👽 👽 👽

[lgfa29]

Minor comments, but it seems to be working 🎉